For decades, moderate drinking appeared protective. The more recent Mendelian randomisation data tell a different story.
The J-curve hypothesis — that moderate alcohol consumption is protective against cardiovascular disease — was one of the most influential ideas in 20th-century epidemiology. Dozens of observational studies throughout the 1980s and 1990s appeared to show that people who drank 1–2 drinks per day had lower cardiovascular mortality than both abstainers and heavy drinkers. This finding was widely publicised, became embedded in public health messaging, and contributed to the cultural normalisation of daily drinking. The 21st-century evidence has substantially undermined it.
The observational studies supporting the J-curve suffered from a well-documented methodological flaw: the 'sick quitter' bias. The abstainer reference group in many studies included former drinkers who had stopped drinking due to illness — making them appear sicker than moderate drinkers. When studies restricted the abstainer group to lifetime never-drinkers, the apparent protective effect of moderate drinking largely disappeared. A 2016 meta-analysis by Holmes et al. in the BMJ, which carefully separated lifetime abstainers from former drinkers, found no significant cardiovascular benefit from moderate drinking.
Mendelian randomisation uses genetic variants as natural experiments to estimate causal effects, bypassing many of the confounders that plague observational studies. Millwood et al. (Lancet, 2019) applied this approach to 512,715 Chinese adults, using genetic variants in ALDH2 and ADH1B that predict alcohol metabolism and consumption. They found a linear, dose-dependent relationship between genetically predicted alcohol consumption and cardiovascular disease — no J-curve. The GBD 2016 Alcohol Collaborators analysis of 195 countries and 28 million people concluded that the safest level of alcohol consumption for overall health is zero drinks per day. The risk of overall harm increases from the first drink.
The IARC classifies alcohol as a Group 1 carcinogen — the highest certainty category, meaning there is sufficient evidence of causality in humans. Alcohol is causally linked to at least 7 cancer types: mouth, pharynx, larynx, oesophagus, liver, colorectum, and female breast. The mechanism is primarily acetaldehyde, the first metabolite of ethanol, which is a direct DNA-damaging agent that forms adducts with DNA and interferes with DNA repair. Acetaldehyde is classified as a Group 1 carcinogen in its own right. There is no safe threshold for cancer risk: even light drinking (1 drink per day) is associated with a small but measurable increase in breast cancer risk.
Beyond cancer, alcohol harms longevity through multiple mechanisms. (1) Epigenetic aging: a 2020 study in Molecular Psychiatry found that heavy drinkers had epigenetic ages 2–3 years older than matched non-drinkers on the GrimAge clock, which is the most predictive clock for mortality. (2) Sleep architecture disruption: alcohol is a sedative that reduces sleep latency but fragments sleep architecture, suppressing REM sleep and increasing slow-wave sleep in the first half of the night while causing rebound wakefulness and REM in the second half. Even moderate drinking (2 drinks) reduces sleep quality by approximately 24%. (3) Liver disease: alcohol is the leading cause of liver cirrhosis in high-income countries. The liver processes approximately 90% of ingested alcohol, and chronic exposure leads to fatty liver, alcoholic hepatitis, and cirrhosis. (4) Neurodegeneration: alcohol is directly neurotoxic. Heavy drinking is associated with brain volume loss, particularly in the prefrontal cortex and hippocampus. A 2017 study in the BMJ found that even moderate drinking (14–21 units/week) was associated with hippocampal atrophy and cognitive decline over 30 years.
The GBD 2016 analysis estimated the following all-cause mortality risk increases relative to non-drinkers: 0.5 drinks/day: +0.5% relative risk. 1 drink/day: +0.7%. 2 drinks/day: +7%. 5 drinks/day: +37%. These numbers appear small at low doses, but they accumulate over a lifetime. Heavy drinking (>14 units/week for women, >21 for men) is associated with a 4–5 year reduction in life expectancy. Alcohol use disorder is associated with a 20–30 year reduction in life expectancy.
The most honest answer is: probably not, and certainly not enough to recommend drinking for health. The apparent cardioprotective effect of moderate drinking in observational studies is most likely explained by confounding and the sick quitter bias. Even if a small cardioprotective effect exists for ischaemic heart disease, it is more than offset by increased risks of haemorrhagic stroke, cardiomyopathy, atrial fibrillation, and the seven alcohol-attributable cancers. The net effect on all-cause mortality is neutral to harmful even at moderate doses.
For those who choose to drink, the evidence supports the following harm-reduction strategies: (1) Keep consumption below 7 units/week (the level at which all-cause mortality risk begins to rise meaningfully). (2) Avoid drinking on consecutive days — alcohol-free days allow the liver to recover. (3) Never drink within 3 hours of bedtime — this is when alcohol most severely disrupts sleep architecture. (4) Avoid drinking before exercise — alcohol impairs muscle protein synthesis and recovery. (5) If you have a family history of breast cancer, colorectal cancer, or liver disease, the risk-benefit calculation shifts further toward abstinence. (6) Red wine is not meaningfully different from other alcoholic beverages in terms of health risk — the resveratrol content is too low to have measurable biological effects at typical drinking doses.
Vitaei verdict
The evidence no longer supports a protective effect of moderate alcohol consumption. The safest level for longevity is zero or as close to zero as is practical. The J-curve was a methodological artefact. Alcohol is a Group 1 carcinogen, an epigenetic aging accelerant, and a sleep disruptor. If you choose to drink, keep it below 7 units/week, avoid consecutive days, and never drink within 3 hours of sleep.
Based on Wood et al. (Lancet, 2018) — 599,912 current drinkers across 83 prospective studies. "Drinks" = standard UK units (8 g ethanol). US standard drink ≈ 1.75 UK units.
Sources: Wood AM et al. "Risk thresholds for alcohol consumption." Lancet 2018;391:1513–1523. doi:10.1016/S0140-6736(18)30134-X. GBD 2016 Alcohol Collaborators. Lancet 2018;392:1015–1035. Estimates are population medians; individual outcomes vary.